Table 1.
Summary of Resampling Results in Training Dataset for ColSBE Using RT-PCR Values
| Ct standardized | |||
|---|---|---|---|
| Estimate | Training | Test | All |
| Sensitivity | 89.0 | 80.3 | 86.0 |
| Specificity | 83.8 | 80.7 | 84.1 |
| Accuracy | 85.8 | 80.7 | 84.1 |
| % IDH1 mut GLE | 13.3 | 7.7 | 10.6 |
| % IDH1 mut GHE | 0 | 0 | 0 |
| Fold-change GHE/GLE MGMT | 2.1 | 2.3 | 2.0 |
| % GLE <mean MGMT | 59.6 | 65.1 | 62.2 |
| % GHE <mean MGMT | 33.3 | 39.0 | 35.5 |
| Ct quantiles normalized+standardized | |||
|---|---|---|---|
| Estimate | Training | Test | All |
| Sensitivity | 97.0 | 86.4 | 93.3 |
| Specificity | 83.7 | 81.6 | 83.0 |
| Accuracy | 88.9 | 83.6 | 87.2 |
| % IDH1 mut GLE | 12. | 7.4 | 10.0 |
| % IDH1 mut GHE | 0.0 | 0.0 | 0.0 |
| Fold-change GHE/GLE MGMT | 1.8 | 1.8 | 1.6 |
| % GLE <mean MGMT | 58.5 | 63.9 | 61.1 |
| % GHE <mean MGMT | 33.4 | 38.3 | 35.2 |
This table depicts the average classification and molecular features across iterations based on the stratification resulting from LDA equations fitted with the RT-PCR values from the ColSBE. On the top panel, Ct values were transformed to a zero centred distribution by subtracting the Ct value of a sample from the mean of all samples for a given gene and divided by the standard deviation (i.e., for a gene 1 and a sample 1 the computation would be [Mean Ct gene1 – Ct gene1-sample1]/standard deviation Ct gene1). On the bottom panel, Ct values were first normalized by the quantiles method and the same standardization than the one described above was undertaken. The 42 cases composing the training set were subjected to an iterative process that was repeated 10,000 times. Such set of samples was split in a further training (2/3 of cases) and test (1/3 of cases) set at each iteration. The training set was used to develop the LDA equation and the obtained discriminant coefficients were multiplied by expression values from the test set, which resulted into a single discriminant score per sample. Those cases displaying a negative score were classified as GLE, while as GHE those ones showing a positive one. The sensitivity, specificity, and accuracy (see Methods) were computed taking as a “gold standard” reference the classification obtained by the ColSBE from gene expression microarrays. Also, the percentage of cases harboring the IDH1 mutation per group is described (% IDH1 mut GLE or GHE), the MGMT fold-change GHE/GLE and the percentage of cases per group having a MGMT expression value below the average of all cases (% GLE or GHE <mean MGMT).